Word Recognition in Two Languages and Orthographies: English and Greek

Home , Greek orthography, PRO (linguistics)

Memory & Cognition 1994, 22 (3), 313-325 Word recognition in two languages and orthographies: English and Greek

HELENA-FIVI CHITIRI St. Francis Xavier University, Antigonish, Nova Scotia, Canada and DALE M. WILLOWS Ontario Institute for Studies in Education, Toronto, Ontario, Canada

Word recognition processes of monolingual readers of English and of Greek were examined with respect to the orthographic and syntactic characteristics of each language. Because of Greek's direct letter-to-sound correspondence, which is unlike the indirect representation of English, the possibility was raised of a greater influence of the phonological code in Greek word recognition. Because Greek is an inflected language, whereas English is a word order language, it was also possible that syntax might influence word recognition patterns in the two languages differen tially. These cross-linguistic research questions were investigated within the context of a letter cancellation paradigm. The results provide evidence that readers are sensitive to both the ortho graphic and the linguistic idiosyncracies of their language. The results are discussed in terms of the orthographic depth hypothesis and the competition model.

In this research, we examined word recognition in En ber (singular, plural), and the case (nominative, genitive, glish and Greek, two languages with different orthogra accusative, vocative) of the noun. Verb endings provide phies and syntactic characteristics. We examined the rela information about person and number as well as about tionship of orthography to the role of the phonological the voice and the tense of a verb. code in word recognition as it is exemplified in the pro Greek words are typically polysyllabic (Mirambel, cessing of syllabic and stress information. To study the 1959). Two-syllable words are most common, followed role ofthe syntactic system, we investigated whether word closely by three-syllable words; four- and five-syllable recognition patterns in the two languages would differ words occur much less frequently. Function words are when readers were processing function words and the dif usually shorter than content words, having only one or ferent parts of content words. Of special importance in two syllables. Most syllables in Greek are open Greek, an inflected language, was whether readers would (Magoulas, 1979) with the predominant pattern being a pay more attention to inflected function words and to the consonant-vowel syllable. parts of content words carrying inflections. Greek stress does not have a fixed position: it can be placed on any of the last three syllables of a word, de Greek Language and Orthography pending on phonological, morphological, and lexical fac An Indo-European language, Greek depends primarily tors (Tombaides, 1986). Stressed syllables are somewhat on inflections to denote semantic and syntactic relation louder and longer than unstressed syllables, although the ships. In modem Greek, the noun, its modifiers, and the difference is not as pronounced as in English. Most Greek verb carry inflections. Noun inflections give information words have a single stress, which in written language is about the gender (masculine, feminine, neuter), the num- indicated by a stress mark over the appropriate syllable. A secondary stress often develops in polysyllabic words when the stressed syllable precedes or follows several un stressed syllables (Joseph & Philippaki-Warburton, 1987). The authors acknowledge their gratitude to the students who partici The 24-letter alphabet of modem Greek represents 26 pated in this study and to their teachers and school principals. For the phonemes. Sounds are represented directly in the orthog preparation of this article we wish to thank the editor, Margaret Jean raphy because ofthe consistency ofthe letter-to-sound cor Intons-Peterson; Michael Turvey; and two other anonymous reviewers for their constructive and helpful comments. For their generous editorial respondence, in most cases between a phoneme and a sin help we want to thank our colleagues Gary Brooks and Ken den Heyer gle letter. Orthographic variation is permitted for some and our friend Wenda Young. Correspondence concerning this article vowel sounds, which can be represented by different let should be addressed to H. F. Chitiri, Department of Education, St. ters or letter combinations. However, a word's spelling Francis Xavier University, Antigonish, NS, Canada B2G ICO (e-mail: provides direct information about its pronunciation be [email protected]). cause each letter's single sound remains constant in dif -Accepted by previous editor, Margaret Jean 1ntons-Peterson ferent contexts.

Orthographic System Effects nodes. The extent to which the activated letter and pho The orthographic depth hypothesis (Feldman & Tur neme clusters will correspond to syllabic units depends vey, 1983; Katz & Feldman, 1983) predicts that the not only on the saliency of syllables in the language, but phonological code should playa more important role in also on the transparency of the phonological representa lexical access in shallow orthographies (e.g., Serbo tions of the orthography. In studies ofEnglish word rec Croatian), which are characterized by an isomorphism be ognition, the use ofsyllabic units has been found in lower tween graphemes and phonemes, than in deep orthogra frequency words only (Jared & Seidenberg, 1990; Seiden phies, in which the mapping between spelling and sound berg, 1987, 1989). This tendency seems to be justified is less transparent. This is a dimension on which the or not only because of the lack of a uniform syllabic struc thographies of Greek and English, even though both are ture in English, but also because of the ambiguity in the alphabetic, differ: Greek orthography is characterized by sound representation of words. Conversely, in the shal a more isomorphic relationship between phonemes and low Serbo-Croatian orthography, syllabic units seem to graphemes than is English orthography. be involved in the recognition ofall words (Katz & Feld Research on English has indicated that in skilled word man, 1981). recognition the phonological code has a lesser role than Stress, the other property upon which we focused, gives that of the visual code, mainly because of the complex critical information not only about a word's correct pro letter-sound correspondences of the orthography. This nunciation, but also about the retrieval of its phonologi complexity renders the phonological pathway slow and cal form (Colombo, 1992). This information may be es unreliable (Coltheart, 1978). As a result, some theorists pecially relevant to word identification in languages in suggest that the phonological representation of English which stress location is not predictable (Black & Byng, applies only to the predictable parts ofthe word-mainly, 1986). Stress effects are apparent only when a word's to consonants and consonant clusters (Carr & Pollatsek, phonological form is assembled on the basis of letter 1985), or to lower frequency words only (Seidenberg, sound correspondences (Colombo, 1992). In such in 1985). Other theorists propose that the phonological stances, specific syllables may emerge as more decisive code's role is mainly that of preserving the information and consequently more sensitive to stress location than in short-term memory (addressed phonology; see Patter other syllables. This seems to be true of final syllables son & Coltheart, 1987). in both English (Drewnowski & Healy, 1982; Goldman Research on alphabetic writing systems with a more & Healy, 1985; Treiman, Goswami, & Bruck, 1990) and regular letter-to-sound correspondence than that found in Italian (Colombo, 1992). English has suggested a stronger involvement of the In the present experiment, the orthographic depth hy phonological code in lexical access. For example, in both pothesis led us to expect that syllable and stress effects lexical decision and naming tasks, Serbo-Croatian readers would be more pronounced in Greek than in English word are influenced more by the phonological forms of the recognition because of the Greek orthography's con words (Lukatela, Savic, Ognjenovic, & Turvey, 1978; sistency in representing the sounds of the language. Lukatela & Turvey, 1980, 1990), and less by qualities operating on the lexical level, such as semantic priming Language System Effects (Frost, Katz, & Bentin, 1987) or frequency (Feldman & The second aspect of the reading process that we ex Turvey, 1983). Research has likewise indicated a more amined was the processing of syntactic information-in pronounced role ofthe phonological code in word recog particular readers' sensitivity to the syntactic cues avail nition in German than in English (Scheerer, 1987). able in their language. We based our hypothesis on the To study the role ofthe phonological code in word rec competition model (MacWhinney & Bates, 1989; Mac ognition, we examined the syllable and stress. Syllables Whinney, Bates, & Kliegl, 1984), according to which cue are universal properties ofspeech; they playa major role strength (a linguistic structure's frequency and reliabil in the assignment of the phonological rules within the ity in expressing desired functions) plays a major role in word, providing a framework to its phonological form speakers' processing of language. On the basis of this (Selkirk, 1980). The formation of syllables follows the model, linguistic structures might differ in importance guidelines for the permissible vowel-consonant combi across languages. This prediction has been confirmed nations in each language. In alphabetic writing systems, through cross-linguistic studies which have shown that in information about syllables is reflected to some extent in interpreting sentences, speakers utilize different cues in words' spellings because letter combinations between syl accordance with the syntactic characteristics of their lables are less frequent than those within syllables (Adams, respective languages (MacWhinney et al., 1984). 1981). In the area of word recognition, support for the com Recent word recognition models have included syllabic petition model comes from studies on the storage and pro information in both the orthographic and the phonologi cessing ofpolymorphemic words and function words. In cal systems oflexical access (Seidenberg, 1989; Seiden the case ofpolysyllabic and polymorphemic words in En berg & McClelland, 1989). This information is encoded glish, certain researchers have proposed that first sylla in terms of letter and phoneme clusters, which are acti bles enjoy a more prominent status than last syllables vated through the corresponding letter and phoneme (Taft, 1979, 1987; Taft & Forster, 1976). The lesser role WORD RECOGNITION IN ENGLISH AND GREEK 315

ofthe last syllable is supported by studies suggesting that recognition? And second, do readers ofdifferent languages the most effective prime of an inflected word is its root attend to words or parts of words in accordance with the morpheme (Lima & Pollatsek, 1983) and that the use of syntactic characteristics of their respective languages? an inflected prime has the same facilitative influence on The role of the phonological code was studied in rela a base verb as does the use of the base verb itself ( Stan tion to students' sensitivity to syllables and stress in two ners, Neiser, Hernon, & Hall, 1979). Consequently, some and three-syllable content words. On the basis of the or theorists have suggested that inflectional affixes may not thographic depth hypothesis, it was expected that these be fully represented in memory, but may be appended to two phonologically defined features would playa stronger roots through the application of rules. role in Greek than in English. In contrast, research on inflected languages such as The impact of syntactic characteristics was examined Serbo-Croatian and Italian has provided support for the in relation to the processing of the different parts ofcon importance of the last syllable in word recognition. In tent words and of function words. We further examined Serbo-Croatian, subjects respond faster to targets in whether Greek readers would discriminate between in oblique cases when they are primed by an identical prime flected and uninflected function words. On the basis of than when they are primed by another oblique form of previous research, we expected that such readers would the same noun (Feldman, 1987; Feldman & Fowler, 1987; attend particularly to content word endings and to inflected Feldman, Kostic, Lukatela, & Turvey, 1983). Similarly function words. English readers, on the other hand, were in Italian, inflections are an important differentiating factor expected to attend more to content than to function words in reaction time to verbs: subjects' response times vary and to focus primarily on the beginnings ofcontent words. (60%), depending on the number of letters in an inflec The subjects were skilled monolingual English and tion (Jarvella, Job, Sandstrom, & Schreuder, 1987). Greek readers in comparable cities in North America and Differences in the processing of syntactic information Greece. The study was undertaken in two phases. First, across languages have also been found in the case offunc students' linguistic and reading proficiency was assessed tion words. Studies in English have suggested that readers with a battery of pretest measures. Next, these readers' pay less attention to function words than to content words. word recognition processes were investigated with a let A common finding is that function words, especially the ter cancellation paradigm. Both phases were conducted shorter ones such as the, of, and and, are more likely to by the same investigator. The procedures were identical be skipped than content words ofcomparable length (Just for both student populations, and every effort was made & Carpenter, 1980; O'Reagan, 1979). When the letter to ensure comparability of materials and instructions. cancellation paradigm is employed, researchers have found that readers omit more letters in function words than METHOD in content words (Corcoran, 1966; Hatch, Polin, & Part, 1974; Healy, 1976). Subjects Research on the reading of inflected languages, how A total of 137 monolingual subjects participated in this experi ever, has shown that the attention paid by readers to func ment, 72 English- and 65 Greek-speaking students. All attended Grade 10 in public schools and came from middle-class, suburban tion words is the same as, or even more than, that which areas in North America or Greece. The average age was 15 years they pay to content words. Clarke (1979) found that, in (age range, 15-17 years). Volunteer participants were given ques an oral reading task, Spanish readers produced most read tionnaires on their language use patterns. Only students who spoke ing errors on function words. In German, readers ex the respective languages both within and outside the home, and who hibited the tendency to fixate longer on function words participated in all sessions, were included in the sample. The final than on the following content word, a tendency that ap sample consisted of 105 students, including 53 English (33 male and 20 female) and 52 Greek (26 male and 26 female) students. plied especially to inflected function words such as some All were average to above average readers. To ascertain the stu forms ofthe definite article (e.g., der, den; see Bernhardt, dent subjects' linguistic and reading proficiency, they were given 1986). a number oftest measures. The linguistic measures tested vocabulary In accordance with the competition model, we expected knowledge (Peabody Picture Vocabulary Test-Revised, PPVT-R; that readers ofEnglish and Greek would attend to differ Dunn & Dunn, 1981) and syntactic knowledge (Geva & Ryan, ent cues. In particular, Greek readers were expected to 1987). The students' reading proficiency was tested in relation to their context-free word recognition (Wide Range Achievement attend to a greater extent than would English readers to Test-Revised, WRAT-R, Reading; Jastak& Wilkinson, 1984) and the processing of inflections in both content and func their reading comprehension (Stanford Test of Academic Skills, tion words, because of the inflected nature of the Greek TASK, Levell-A-Reading; Gardner, Callis, Merwin, & Madden, language. 1972).1 The general purpose of the present experiment was to The measures were given in the subjects' respective languages. study the effect ofwriting and language system character Only the test of syntactic knowledge (Geva & Ryan, 1987) was avail istics on the reading process. Two main research questions able in Greek and had been used previously in research (Katsaiti, 1986). The other tests had to be translated into Greek. The transla were addressed: Is the consistency of the orthography in tions were done by a native speaker of Greek who was also highly representing the sound of the language a determining proficient in English; they were subsequently checked and approved factor in the involvement ofthe phonological code in word by Greek high school teachers.2 316 CHITIRI AND WILLOWS

Psychometric Issues specific target letter that they notice while they are reading con On the Reading Comprehension subtest of the widely used Stan nected text. The underlying assumption for this technique is that ford Achievement Test (maximum raw score, 78), the English stu readers detect the letters that are processed to a conscious level and dents (M = 58.66, SD = 7.68) and Greek students (M = 59.54, miss those that are not. Moreover, target omission is considered SD = 6.44) performed at comparable levels [F(I, 103) < I]. Per to take place more readily when readers form internal representa formance on this test was almost identical for the two groups, and tions in the form ofunits larger than the letter (Healy & Drewnowski, neither group's performance was near ceiling. All subjects read with 1983). In other words, omission patterns are considered to reflect comprehension at or above the level that would be expected for word processing factors (Proctor & Healy, 1985). their age and grade, so we concluded that both groups consisted The subjects were given normal printed texts and were explicitly of competent readers. This observation was confirmed both by their instructed to focus on comprehension and to cross out target letters teachers and by our own experience in testing them in the ex when they noticed them. As in previous research (Proctor & Healy, periment. 1985), this precaution was taken to ensure that readers processed Our pretesting did, however, raise some serious psychometric the meaning ofthe texts rather than simply scanned the texts to de issues that are implicit in most cross-linguistic reading studies where tect letters. there are no standardized tests for one or more of the language groups. It was apparent from our pretesting of the students that, Materials because of linguistic and orthographic differences between the lan The experimental questions were studied in relation to two pas guages, "matching" groups of English and Greek subjects on the sages, each with a different target letter. The letters chosen, a and basis of linguistic and word recognition test scores is impossible 0, have similar relative frequencies in the two languages. (The let without tests that have been standardized specifically for each lin ter a is the most frequent and 0 the third most frequent vowel in guistic group. This problem was apparent in the translation of the English, according to Massaro, Taylor, Venezky, Jastrzembski, & PPVT-R. Because some ofthe difficult words in English had Greek Lucas, 1980, and also in Greek, according to a preliminary count origins, they were more familiar to Greek than to English students. of vowels of words from a Grade 10 reader.) A significant discrepancy in test performance on the PPVT-R (max Two texts were constructed in each language. All the texts were imum raw score, 175) between the two groups [F(l, 103) = 93.56, biographies. They were written by the same author and approved p < .001], with the English students (M = 141.53, SD = 8.10) ofby high school teachers in each country. The texts had approxi performing considerably below the Greek (M = 157.00, SD = mately the same lengths (2,100 words) and had similar numbers 8.29), is probably at least as indicative of problems related to es of sentences (mean, 80; range, 78-81) and words per sentence tablishing a comparable difficulty level between languages as it is (mean, 26; range, 25.9-26.1). The target letter in each text was ofany real differences in linguistic abilities. On the test of syntac contained in a controlled number ofwords: 135 content words and tic knowledge (Geva & Ryan, 1987; Katsaiti, 1986; maximum raw approximately 300 function words per passage. Examples of the score, 30), the Greek students (M = 27.50, SD = 2.95) also per text materials are given in the following excerpts (see Figure I) formed somewhat better than the English students (M = 25.28, with the content and function words containing the target underlined. SD = 2.74) [F(I,103) = 15.90, p < .001], despite the fact that The content words included two- and three-syllable words that performance on this test in both languages was near ceiling levels. could have the target letter in any syllable (e.g., carpenter, occa This may reflect, in part, a difference across languages in the rela sion, registrar). For each syllable position, there was a stressed tive difficulty ofthe vocabulary in the test. Although the Greek ver and an unstressed condition (e.g., taxi, lament).3 This manipula sion of the test had been used once in previous research, neither tion allowed for the study ofphonological effects among two- and it nor the English version of the test had undergone standardiza three-syllable words across the different syllables and permitted the tion procedures. control ofeffects that are primarily due to target location. The only A different type ofproblem affected the Greek measure ofcontext exception to this pattern was the third syllable of three-syllable free word recognition, which was a translation of the WRAT-R words, which was present in the unstressed condition only. The Reading subtest (maximum raw score, 89). We obtained a very large reason for this restriction was that not enough English words met discrepancy between the English (M = 58.62, SD = 6.89) and these two criteria (third syllable and stressed) for the target letters Greek students (M = 88.48, SD = 0.89) on the WRAT-R under consideration. As a result, there were 135 content words [F(1,103) = 959.69,p < .001], with the Greek students performing divided into nine categories of 15 words each. almost at ceiling. Because of the shallow nature of Greek orthog raphy, the students were able to "read" virtually every word by sounding it out, even if they had no idea what the word meant (as was very clear when they were informally asked word meanings). Such a leader was Sam Martin, the big newspaper For readers ofEnglish, performance dropped drastically once words editor, who was seeking !2 be a nominee of the ceased being familiar. This contrast in the context-free word rec Republican Party for the presidential race in Novem ognition performance across the two languages simply confirms the ber. Sam appeared to be an ideal candidate as he distinction between deep and shallow orthographies and points to seemed !2 embody those virtues that many would like a problem inherent in any comparison between the two types of orthographies: it is possible for a reader of a shallow orthography to see in a public figure, and besides, nobody could to decode virtually any word on the basis of letter-to-sound cor accuse him of being dishonest. respondences. Although every effort was made to equate the samples on the ba sis of linguistic and word recognition measures, we learned "the ETO X€PTeO TWP apaJLP~UfWP TT/S aVTwp fLPCiL hard way" about problems inherent in such attempts. We have de 7raPTa Ot :Y0PfLS TT/S, "(ta TOVS 07rOLOVS 'T/ MaeLa scribed these problems here to alert future researchers to possible difficulties in cross-linguistic research, in the hope that this aware iTeftpf a7r€eapTT/ fXTLJL'T/U'T/ XCiL Ot 07rOLOt fLxap ness will lead to potential solutions. f7rteeroUft {3a8t&. TT/ rwiJ TT/S JLf TO 7raeaoft"(JLCi XCiL ns UVUTaUftS TOVS. Task The task used in the research was the letter cancellation para Figure 1. Excerpt from the English and Greek texts with the tar digm. In this task, readers have to cross out every instance of a get letter o. WORD RECOGNITION IN ENGLISH AND GREEK 317

Words were matched between languages in relation to numbers and target position within each language separately. Moreover, the ofletters, so that each English word was paired with a Greek word omission rate among content words was studied with respect to word having the same letter length (4-12 letters). Among three-syllable frequency (low, medium, high), word length, and the position of words, the majority (71 %) were 7 or 8 letters long, and among the word in the sentence. two-syllable words, most (78%) were 5 or 6 letters long. An at tempt was also made to equate these words in terms of frequency. Since no frequency lists are available in Greek, the frequency of RESULTS the English translation ofeach word was used. Words were classi fied as low (1-40), medium (41-120), and high frequency (121-472) The results of the test phase are based on analyses of on the basis of the frequency list of Ku(!era and Francis (1967). perfonnance on the letter cancellation paradigm. Accord Low-frequency words constituted 65%, medium-frequency words ing to this methodology, detection of a letter is seen as 18%, and high-frequency words 12% of the words used. Every an indication that the letter has been processed to the level Greek word was matched with an English word with exactly the ofconsciousness (Healy, Conboy, & Drewnowski, 1987). same characteristics in terms of number of syllables, target posi tion, stress condition of the syllable containing the target, letter Alternatively, omission ofa letter suggests the formation length, and approximate frequency. of units larger than the letter. Performance on the letter Finally, all texts were followed by six comprehension questions cancellation task in connection with overall omission and in a multiple-choice format. Students who failed to answer three the specific experimental questions studied were analyzed or more out of the six questions correctly were excluded from the by using analyses of variance (ANOVAs) and planned sample. comparisons. We used Pearson correlations to examine Procedure the relationship between letter cancellation on the 0 and The subjects were trained and tested on the experimental task in the a text in each language and perfonnance on the com three different sessions in their normal classrooms. The first ses prehension questions following each passage. Finally, we sion was a brief training session in which the subjects were famil used regression analyses to study the effects of word fre iarized with letter cancellation and with the instructions. In this prac quency, word length, and word position on the omission tice session, in order to avoid any transfer effects, the letter i was of targets in content words. used. The materials used for this training session were three short paragraphs, each of which was followed by comprehension ques tions in a multiple-choice format. In the actual sessions as well, Letter Detection the experimental passage was preceded by a short training para Only misses (failure to detect the target letter) were graph with the same letter as that used in the actual text; this train counted as omissions, because false alarms (cancellations ing paragraph too was followed by multiple-choice questions. The of nontarget letters) were virtually nonexistent. We instructions given in each session were the same and were similar counted omissions in a word only once even in the case to those employed by other researchers (e.g., Goldman & Healy, ofsome function words that included two instances ofthe 1985); the students were asked to focus on comprehension and to cross out every instance ofthe target whenever they noticed it. They letter under consideration (against for the letter a). Sub were told further that they should not go back if they realized that jects who failed to detect any letters on a whole page or they had missed crossing out a letter. During each testing session, who detected only a minimal number ofletters (fewer than a time limit was set on the students' performance, to prevent them the established criterion of42% of content words) were from focusing primarily on letter cancellation. This limit was based excluded from the sample. This was the case for 11 sub on extensive pilot testing, with students' speed set at a minimum jects or 16% from the original North American sample of 105-110 words per minute. Students who performed below this limit were excluded from the final sample. and 12 subjects or 18% from the original Greek sample. The exclusion ofthese subjects was considered necessary, Design because they were not following the task instructions ex The experiment had three major independent variables: language actly (reading for meaning and simultaneously crossing (English, Greek), target letter (0, a), and word class (content or out targets). function). The dependent measure was the omission rate-that is, Target omission was analyzed in an ANOVA in rela the number of instances of the target letter that a subject failed to detect. Language (English, Greek) was a between-subjects factor, tion to the effects of language (English and Greek) and while letter (0, a) and word class (content or function) were the target letter (a, 0). Both main effects were significant. main within-subjects factors. Overall, the major analysis in rela Greek students missed more targets overall than did tion to word class involved a language x letter x word class or English-speaking students [F(I,103) = 5.89, p < .05]. a 2 x 2 x 2 mixed design. Both groups missed more letters when the target was a In addition, we analyzed the omission rate among content words than when it was 0 [F(l,103) = 65.60,p < .001], sug only in relation to three additional independent variables: number gesting an influence of letter frequency, since a occurs ofsyllables (two and three), presence or absence ofstress, and po sition ofthe target (first, second, or third syllable). In this analysis more frequently than 0 in both orthographies. Moreover, as well, language (English, Greek) was a between-subjects factor, the differential omission rate between the two targets was and the letter under investigation (a, 0), the number of syllables more pronounced among Greek students, as is suggested (two or three), the presence ofstress (present, absent), and the po by the significant language X letter interaction [F(I,103) = sition of the target (first, second, or third syllable) were within 4.~5, P < .05]. The higher omission rate in the a pas subjects factors. Overall, the analysis of these three attributes in sage in Greek could be explained by the higher incidence relation to the omission rate among content words involved a lan guage x letter x number of syllables x stress x target position ofmissed function words. Since more noninflected func or a 2 x 2 x 2 x 2 x 3 incomplete crossed design. We used Bon tion words had the letter a (67%) than the letter 0 (48%) ferroni t tests to examine the effects of number ofsyllables, stress, in Greek, the higher omission rate could indicate that they 318 CHITIRI AND WILLOWS were processed as units, which is consistent with the pre 35 vious literature on function words. c: 0 30 'iii 0 1st pos- III Effects of Orthographic System Characteristics 'E 25 stressed Information about the role of the orthography was pro 0 ~ 1st pos- 20 vided by the test of context-free word recognition given 0 unstressed prior to the experimental phase. Students' performance ~ ~ 2nd pos- 0 15 on the WRAT-R Reading subtest (Jastak & Wilkinson, c: stressed as 10 1984) was very different in English than it was in Greek. Q) ~ 2nd pos- The Greek students pronounced almost all words correctly E 5 unstressed (their mean accuracy was 88.48 out of89 words), whereas o English students experienced difficulties once the words English Greek ceased being familiar. In other words, the Greek students Language confirmed the shallow nature of Greek orthography by their perfect word recognition performance, even decod Figure 3. Omission pattern in relation to stress and syllable ef ing words whose meaning was totally unknown to them. fects in two-syllable content words in English and Greek. The first main experimental question regarded the ef fect ofthe orthographic code, in pursuit ofthe prediction of the orthographic depth hypothesis that the degree of the processing oflonger words with more syllables. En phonological representation should be reflected in a glish students exhibited the reverse pattern, since they stronger influence of the phonological code in word rec failed to detect more target letters in two- as opposed to ognition. To investigate the involvement of the phono three-syllable words [t(52) = -3.98,p < .001]. The tar logical code, we focused on two phonologically defined get letter, whether a or 0, did not have an effect on the factors: syllables and stress. These two factors are closely omission pattern in relation to the number of syllables. related, since the presence of stress effects is dependent Since the number ofsyllables and ofletters was controlled upon the formation of reading units larger than the letter in each language, it seems that, in support of the ortho (Drewnowski & Healy, 1977). graphic depth hypothesis, syllabic units do playa stronger Processing of syllabic units. To examine the role of role in Greek word recognition than in English. syllables, we focused on the first and second syllables of Role ofstress. The results ofthe analysis of stress ef two- and three-syllable content words. The results of the fects showed that readers in both languages are influenced analysis indicate that both groups of subjects were influ by their presence, as is indicated by the significant main enced by the number of syllables in a word, which is effect of stress [F(I,103) = 29.02, p < .001]. The ef reflected in a significant main effect ofthe number ofsyl fects of stress, however, were not uniform across two lables [F(1,103) = 6.78, p < .05]. The importance of and three-syllable words, as is indicated by the stress X syllabic units, however, seems to differ between the two number ofsyllables interaction [F(1, 103) = 16.07, P < languages, as is reflected in a significant language x num .001]. Neither were they uniform between the two sylla ber ofsyllables interaction [F(l ,103) = 56.81,p < .(XU] ble positions, whether first or second, as is suggested by (see Figure 2). The Greek students failed to detect more the stress x syllable position interaction [F(l, 103) = letters in three-syllable words [t(51) = 6.44, p < .001], 20.98, P < .001]. suggesting that their omission pattern was influenced by As a result of the different stress effects in relation to number of syllables, separate analyses were undertaken among two- and three-syllable words. Among two-syllable 30 ~------, words, stress effects were not evenly distributed between the two syllable positions [F(l,103) = 16.70, p < .OOl]. c:: 25 o Further, the pattern ofstress effects on each syllable was 'en not consistent between the two languages, as is indicated .!!2 20 by a significant language x stress X syllable position E interaction [F(1,103) = 17.79,p < .001] (see Figure 3). o D 2 syllables '0 15 That is, in Greek, stress effects were present on both the o ~ 3 syllables first [t(51) = -3.46, p < .001] and the second syllables ~ 10 [t(51) = -2.66, p < .01]. In English, on the other hand, c:: ctl stress effects were limited to the second syllable only Q) 5 E [t(52) = -7.44, p < .001]. This result in English could in part be a consequence of the linguistic nature of the o .1....---'-_ English Greek stimuli; in the present study, English two-syllable words with the target in the stressed second syllable were mainly Language verbs, whereas words with the target in the unstressed Figure 2. Omission pattern in relation to syllable effects in En condition were nouns. Thus, the observed stronger stress glish and (Jreek. effects on the second syllable of English two-syllable WORD RECOGNITION IN ENGLISH AND GREEK 319

35 this tendency applied overall to both languages [F(1,103) < 1], it was affected by the target letter, as g 30 ·00 is indicated by a significant triple interaction involving .!!l 25 0 1st pas- word class, language, and target letter [F(1,103) = 11.05, E stressed P < .001]. ~ 20 ~ 1st pas- Separate analyses further demonstrated that the target o unstressed ~ 15 letter had a differential effect in each language (see Fig 0 ~ 2nd pas- ure 5). In English, readers demonstrated a tendency to at lij 10 stressed tend less to function words regardless oftarget [F(1,5l) = CD ~ E 5 2nd pas- 3.14, n.s.]. In Greek, on the other hand, the omission unstressed pattern between content and function words was influ 0 enced considerably by the target letter [F(1,50) = 51.77, English Greek p < .001]. The Greek students tended to differentiate Language more strongly between content and function words in the case of the letter a [t(5l) -4.19, P .001]. In the Figure 4. Omission pattern in relation to stress and syllable ef = < fects for the first two syllables of three~syllable content words (tar case ofthe letter 0, however, the Greek students did not get letter 0) in English and Greek. differentiate between content and function words. This outcome could be attributed to the fact that 52 % offunc tion words with the letter ° were inflected, so that they words could also indicate that the last syllable of verbs carried important linguistic information about the follow is attended to more than may be the case for the last syl ing noun. As a result, readers attended more to them, lable of nouns. missing fewer letters. For three-syllable words, stress was not significant On the basis of these findings, it would seem that [F(1,103) = 2, n.s.]. This absence of stress effects ap readers are indeed influenced by the syntactic characteris plied to both languages, even though, as a marginally sig tics oftheir language. In English, readers tend to pay less nificant interaction between language and stress suggested attention to function words overall, whereas Greek readers [F(1,103) = 3.68, p < .058], stress effects were some display this pattern only when the function words are not what more pronounced in Greek. This tendency in Greek inflected. In the case of inflected function words, how was attributed mainly to words having the target 0, in ever, Greek readers seem to attend to them to the same which case students distinguished between the stressed and extent as they do to content words. the unstressed condition in the second syllable, as is sug The processing ofthe last syllable ofcontent words. gested by the significant stress X position interaction Syllable position was examined separately for two- and [F(1,103) = 7.40, P < .05] (see Figure 4). three-syllable words. In both instances, syllable position On the whole, therefore, it can be said that although had a differential effect in each language. Among two readers appear to be sensitive to stress in both languages, syllable words, this effect was indicated by the signifi stress effects seem to be somewhat more prevalent in cant language x syllable position interaction [F(1, 103) = Greek than in English. Even though this conclusion ap 23.82, P < .001]. The Greek students displayed a con- pears to support the orthographic depth hypothesis, the fact that stress effects did not apply generally in Greek qualifies its generality. 40 Effects of Language System Characteristics § 35 The second major research question regarded the im 'w pact ofthe language system's characteristics on the read .!:!? 30 ing process, to assess the extent to which readers demon g 25 ri- conlent strate sensitivity to the syntactic cues available in their rll O words language. In particular, we sought to investigate the pre -o 20 rI- ~ diction of the competition model (MacWhinney et al., 0 15 f ~ function 1984) that cue strength (a linguistic structure's frequency c ~ words ctl 10 and reliability in expressing desired functions) plays a Q) major role in speakers' processing oflanguage. We there E 5 fore examined the processing of content versus function o -- _ ..... c-..... _ ..... words, as well as the processing of the last syllable of o-Eng a-Eng o-Gre a-Gre content words, which carries inflections in both languages. The effect ofword class. The results indicated that stu Language dents attended more to the target letters in content words Figure 5. Omission pattern in relation to the processing of con and less to those in function words [F(l,103) = 11.73, tent and function words in English and Greek for target letters a P < .001]. Even though, consistent with previous studies, and o. 320 CHITIRI AND WILLOWS

35 ,------, tend more to the beginning parts of the word. In Greek, however, readers seemed to attend more to the last sylla 5 30 'u; ble that carried the inflection; this was consistent with rn findings for other inflected languages. 'E 25 o o 20 o 1st position Effects of Word Frequency, Word Length, and ~ 15 Position in the Sentence ~ 2nd position c: Target omission was further analyzed in relation to con ~ 10 ~ 3rd position tent word frequency, word length, and the position of the E 5 word in the sentence. The importance of the last variable was suggested in a letter cancellation study by Smith and Groat (1979), who found that the sentence position ofthe English Greek word containing the target accounted for a substantial per Language centage of omissions. Figure 6. Omission pattern for the three unstressed syllables of The results of a regression analysis of word length in three-syllable content words in English and Greek. relation to these three attributes as predictors indicated that the omission rate was indeed influenced by some of the variables studied in English [F(3,262) = 5.21, p < sistent pattern of attending more to the second syllable, .005] as well as in Greek [F(3,264) = 5.38, p < .005]. which carried the inflection. This effect ofattending more Specifically, in both languages these variables accounted to the second syllable applied both when this syllable was for 6% of the variance. In English, the only variable to stressed [t(51) = 3.15, p < .005] and when it was un exert a significant effect on target omission was word fre stressed [t(51) = 3.14, p < .005]. The English-speaking quency [F(3,262) = 13.08, p < .001], with targets in students, on the other hand, differentiated between sylla frequent words being missed more than those in less fre ble position, depending upon the presence ofstress. When quent words. In Greek, on the other hand, the variable the two syllables were both stressed, the subjects attended to account for most of the variance was word length more to the second syllable [t(52) = 2.48, p < .05]; [F(3,264) = 11.44, P < .001], with more targets being when they were unstressed, they attended more to the first missed in longer words. Moreover, the effects were of syllable [t(52) = -5.55, p < .001]. similar magnitude for the 0 and a passages in both lan Among three-syllable words, the effects ofsyllable po guages. sition were studied in the unstressed condition only, since The effect ofword frequency in English could indicate no three-syllable words had the target in the stressed third that frequent words are recognized faster and on the ba syllable. The effect of syllable position among the three sis of fewer features than are less frequent words. Fur unstressed syllables of three-syllable words was signifi thermore, their recognition does not require the conscious cant both overall [F(2,206) = 5.73, p < .005], and be identification of component letters (Healy et al., 1987). tween languages, as is indicated by the significant inter The absence of frequency effects in Greek could be re action between syllable position and language [F(1, 103) = lated to a possible discrepancy between the actual fre 30.59, p < .001] (see Figure 6). Further comparisons quency ofGreek words and the frequency assigned in this between the three syllable positions indicated that the study. However, there is also the possibility that the lack Greek- and the English-speaking students differed mainly ofa word frequency effect could be real, because studies in the relationship between the third syllable and the first of the shallow Serbo-Croatian orthography have shown [F(1,103) = 39.02, P < .001], and between the third syl that naming in this language is affected less by word fre lable and the second [F(I,103) = 43.98,p < .001]. For quency than is the case in English (Katz & Feldman, the English-speaking students, the third syllable was the 1983). one to be processed the least, exhibiting the highest omis sion rate. This finding applied to the third syllable in re DISCUSSION lation to the first [t(52) = -7.13, p < .001] as well as to the second syllable [t(52) = -6.73, p < .001]. For In the present study, we set out to examine the nature the Greek students, on the other hand, target letters in of the reading process across two different languages and the third syllable, which carried the inflection, were orthographies, in an attempt to gain some insight into the missed somewhat less often than those in the first [t(51) = impact that these two dimensions have on reading pro 2.10, p < .05] and second [t(51) = 2.96, p < .005]. cesses. We studied the influence ofthe orthographic sys The results ofthis analysis as well seem to confirm the tem in relation to the involvement of two phonological hypothesis that readers attend more to the parts of the aspects of word recognition: syllable and stress. We ex word that carry important syntactic information. As in amined the effect ofthe linguistic system with respect to previous findings, English students had a tendency to at- the processing of words in relation to their word class WORD RECOGNITION IN ENGLISH AND GREEK 321

(content or function) and to the last syllable of content In shallow orthographies, such connections are strong be words in comparison with the other syllables. cause the consistency ofthe letter-sound correspondences The evidence relating to these specific questions indi allows for the quick computation ofthe phonological code cated that, contrary to previous suggestions, the reading in parallel with the orthographic code. Moreover, the process is not uniform across languages and orthogra present findings on the use ofsyllabic units in Greek word phies. At least at the levels examined, it reflects the sen recognition are in line with similar findings in Serbo sitivity of the readers to the idiosyncrasies oftheir orthog Croatian (Katz & Feldman, 1981), in which the phono raphy and language. This pattern of results is the more logical code is supposed to be the main route for lexical significant because it was found in skilled readers in each access (Feldman, 1987; Feldman & Turvey, 1983; Frost language, a stage at which differences should have been & Katz, 1989; Katz & Feldman, 1983; Lukatela et al., at a minimum. 1978; Lukatela & Turvey, 1990; Turvey, Feldman, & Lukatela, 1984). The Effect of Orthographic Characteristics In agreement with previous suggestions (Frederiksen on Word Recognition & Kroll, 1976), the confinement in English of syllabic Both the visual and the phonological codes are impli and stress effects to only the last syllable of two-syllable cated in the word recognition of English and Greek. The words seems to indicate the predominance of the visual effect of the visual code was indicated by the presence route in word recognition. This conclusion is also sup ofletter frequency effects in the two orthographies (i.e., ported by the finding of frequency effects on target omis a is more frequent in both languages than 0, and it was sion in the actual texts. missed more often), and the involvement of the phono At the same time, the presence ofthe mentioned phono logical code was suggested by readers' sensitivity to syl logical effects also suggests that the visual route is not labic and stress factors. the only one used in English word recognition. Thus, the Despite the presence of both codes in the word recog convergence of syllabic and stress effects on the last syl nition of the two orthographies, the results also suggest lable oftwo-syllable words lends some support to theories that the involvement of the phonological code may be about the presence of phonological encoding in skilled more pronounced in Greek. This possibility was supported reading of English (Rubenstein, Lewis, & Rubenstein, by the more consistent syllabic and stress effects obtained 1971; Spoehr & Smith, 1973, 1975; Van Orden 1987; among Greek words and also by the concomitant pres Van Orden, Johnston, & Hale, 1988). According to ence of word length effects on the omission rate. The Seidenberg and McClelland's (1989) connectionist model, simultaneous presence ofword length and syllabic effects the phonological code is always generated in English word has also been observed in the naming of English low recognition even though it is not always used. The rea frequency words by Jared and Seidenberg (1990), who son for the phonological code's more limited input in word attributed it to the activation of a word's phonological recognition is related to the lack of strong correlations properties. between English orthography and phonology, a lack that The presence of syllabic and stress effects in Greek contributes to weaker weights on the connections between could also be interpreted in terms of visual encoding. For visual and phonological units. one thing, syllables can be processed on the basis of a It is possible, however, that the role of the phonologi word's orthographic redundancy-that is, on the basis of cal code becomes more prevalent as orthographic infor one's knowledge ofthe permissible letter coalitions in the mation increases and constrains phonological ambiguity, word (Adams, 1981; Seidenberg, 1987, 1989). In addi a condition that applies mainly to word endings (Drew tion, the presence of stress effects primarily among two nowski & Healy, 1982). This explanation of the later syllable words could be ascribed to visual processes, since emergence ofthe phonological code in English word rec stress is also marked visually in Greek. In this case, the ognition also seems to be consistent with theories (Kay, absence ofstress effects in the first two syllables ofthree 1987; Patterson & Coltheart, 1987) that support some of syllable words could be explained in terms of the lesser the predictions ofanalogy models (Glushko, 1979). Here processing of the stress mark because of the added load a word's pronunciation is formed by analogy to other of processing three instead of two syllables (Spoehr & words that share the same spelling patterns and especially Smith, 1973). the last three letters with the actual word. In addition, this However, the shallow nature ofthe Greek orthography interpretation of the present results is in agreement with which was demonstrated in the students' very accurate findings on the importance of rime units for the assign performance on the test ofcontext-free word recognition ment of pronunciation in English (Treiman & Chafetz, (WRAT-R)-makes it more likely that these effects were 1987; Treiman & Zukowski, 1988). The restriction ofthe due to phonological encoding. Such a view is consistent phonological code to the last parts ofthe word could also with connectionist models of word recognition such as explain the absence of syllabic and stress effects among Seidenberg and McClelland's (1989), according to which the first two syllables ofthree-syllable words, even though the connections between orthographic and phonological the majority of these words were oflow frequency. This units are determined by the nature of the orthography. absence of phonological effects among lower frequency 322 CHITIRI AND WILLOWS

words did not conform to the proposition that low fre vious findings regarding the importance of word endings quency is a prerequisite for phonological encoding in in the word recognition of inflected languages (Feldman English (Jared & Seidenberg, 1990; Seidenberg, 1985, & Fowler, 1987; Feldman et al., 1983). Moreover, al 1989). Instead, the present findings suggested that phono though the stimuli used in the present experiment were logical disambiguation may be a more important pre on the whole confined to one linguistic class only (nouns), requisite for the use of the phonological code in English, it is likely that similar findings would be observed for a possibility that could explain the later emergence ofthe other linguistic classes, because ofthe consistency ofthe phonological code in word recognition, regardless of fre last syllable in carrying the inflection in Greek. Support quency. for this conjecture comes from a study of Italian, also an It would seem, therefore, that both the visual and the inflected language, which confirmed the importance of phonological codes are involved in the word recognition the last syllable for the word recognition ofverbs as well ofthe two orthographies, thus supporting models in which (Jarvella et aI., 1987). both codes playa role (see, e.g., Carr & Pollatsek, 1985; Readers ofEnglish also exhibited selectivity in the cues Seidenberg & McClelland, 1989). At the same time, how they processed. In the case of content words, students ever, the extent to which the phonological code is used manifested a tendency to process to a greater extent the for lexical access might not be the same in the two or beginning parts ofthe word. Among three-syllable words, thographies. Specifically, as mentioned earlier, if one is this tendency did not apply only to the first syllable, as to judge by the more consistent syllabic and stress effects suggested by Taft (Taft, 1979, 1987; Taft & Forster, in Greek, the use of a phonological code may be more 1976), but also to the second syllable, as observed in other prevalent in this language than in English. A stronger in studies (Andrews, 1986; Lima & Pollatsek, 1983). The volvement ofthe phonological code in the shallow Greek inclination to pay attention to both the first and the sec orthography than in the deep English one is in line with ond syllables of three-syllable words was also supported the predictions ofthe orthographic depth hypothesis (Feld by the lack ofdifferentiation between these two syllables man & Turvey, 1983; Katz & Feldman, 1983). under stressed and unstressed conditions. Among English The indication that the visual code is also involved in two-syllable words, the second syllable may also be in Greek word recognition, as suggested by the presence of volved to a greater extent in word recognition but only letter frequency effects and by the possibility ofthe visual in the case of verbs. Although one could infer that the processing of stress, departs somewhat from the stronger possibility of different processing of nouns and of verbs version of the orthographic depth hypothesis, however. is another reflection of the impact of linguistic charac One reason for the discrepancy between the Greek results teristics on word recognition, more research is needed to and the Serbo-Croatian ones could be the stronger adher substantiate this suggestion. With respect to word class, ence ofthe Serbo-Croatian script to the principle ofone English readers tended to process content words to a to-one letter-to-sound correspondence (Frost et al., 1987). greater extent than they did function words, in agreement The present fmdings for Greek may conform more closely with the findings in previous studies (Hatch et al., 1974; to those obtained previously for other orthographies, such Healy, 1976; Schindler, 1978; Taylor & Taylor, 1983). as the German, that adhere to the principle of represent On the whole, the findings on the impact of language ing the pronunciation of a word while at the same time system characteristics suggest that readers' behavior is preserving its morphological affiliation (Scheerer, 1987). guided by both universal and language-specific criteria. The first aspect was demonstrated in relation to the ten The Effect of Linguistic Characteristics dency to miss target letters in function words more than on the Reading Process in content words in both English and Greek, a trend in The second question-relating to the effect of linguis dicating some universal properties that function words tic system characteristics on the reading process-also share as a class. Such properties could be related to func suggests some differences in readers' processes. In this tion words' linguistic contribution to meaning, which is case, the present results confirmed the prediction of the similar across languages (Taylor & Taylor, 1983), as well competition model that readers are guided by the nature as to their frequency and to their physical appearance of their language to process informative cues more than (brevity). Because ofthe latter two characteristics, it might less informative ones (Just & Carpenter, 1987; Mac be that function words are processed as units (Healy, Whinney & Bates, 1989; MacWhinneyet al., 1984). This 1980; Healy & Drewnowski, 1983) or that they are pro tendency was manifested in both languages. cessed concurrently with the previous content word (Just In Greek, this trend was expressed in the deeper pro & Carpenter, 1980, 1987). cessing of inflected function words, as was the case for The use oflanguage-specific criteria was demonstrated, function words with the target 0, and for noun endings on the other hand, in the Greek readers' differential pro that carried inflectional affixes. Among nouns, the ob cessing of inflected and uninflected function words. As served pattern oflower omission on the last syllable both mentioned previously, when processing inflected function indicated that readers were sensitive to the linguistic in words, the Greek readers attended to them to the same formation conveyed by this syllable and substantiated pre- extent as they did to content words. When the function WORD RECOGNITION IN ENGLISH AND GREEK 323

words were uninflected, on the other hand, these readers COLTHEART, M. (1978). Lexical access in simple reading tasks. In attended to them considerably less than to content words G. Underwood (Ed.), Strategies ofinformation processing (pp. 151 displaying the same pattern as in English. On the basis 216). London: Academic Press. CORCORAN, D. W. (1966). An acoustic factor in letter cancellation. Na of this result, one can infer that not all function words, ture, 210, 658. but rather only inflected ones, are processed at a deeper DREWNOWSKI, A., & HEALY, A. F.(1977). Detection errors on the and level in inflected languages. This suggestion confirms and: Evidence for reading units larger than the word. Memory & Cog Bernhardt's (1986) findings for German. nition, 5, 636-647. DREWNOWSKI, A., & HEALY, A. F. (1982). Phonetic factors in letter Another indication of the use of language-specific cri detection: A reevaluation. Memory & Cognition, 10, 145-154. teria is related to the different patterns exhibited by the DUNN, L. M., & DUNN, L. M. (1981). Peabody Picture Vocabulary two language groups in the processing of content words. Test-Revised. Circle Pines, MN: American Guidance Service. Regarding this tendency, the results provided evidence FELDMAN, L. B. (1987). Phonological and morphological analysis by contrary to Taft's (1979, 1987) generalization about the skilled readers of Serbo-Croatian. In A. Allport, D. G. MacKay, W. Prinz, & G. Scheerer (Eds.), Language perception and produc uniqueness ofthe first syllable or basic orthographic syl tion (pp. 197-209). London: Academic Press. lable structure for word recognition. Thus, although his FELDMAN, L. B., & FOWLER, C. A. (1987). The inflected noun system hypotheses may be relevant to English, they appear not in Serbo-Croatian: Lexical representation ofmorphological structure. to apply to Greek, in which the last syllable also seems Memory & Cognition, IS, 1-12. FELDMAN, L. B., KOSTIC, A., LUKATELA, G., & TURVEY, M. T. (1983). to influence lexical access. In addition, although the pres An evaluation of the "basic orthographic syllabic structure" in a ent findings cannot provide direct evidence concerning phonologically shallow orthography. Psychological Research, 45, the memory representation of words in each language, 55-72. they can nevertheless offer some suggestions. In Greek, FELDMAN, L. B., & TURVEY, M. T. (1983). Word recognition in Serbo for example, as in Serbo-Croatian (Feldman & Fowler, Croatian is phonological analytic. Journal ofExperimental Psychol ogy: Human Perception & Performance, 9, 414-420. 1987; Feldman et al., 1983), it is possible that lexical en FREDERIKSEN, J. R., & KROLL, J. F. (1976). Spelling and sound: Ap tries are fully represented in memory, as is suggested by proaches to the internal lexicon. Journal of Experimental Psychol the importance of the last syllable. If this hypothesis is ogy: Human Perception & Performance, 2, 361-379. true, it conflicts with the proposition that words are rep FROST, R., & KATZ, L. (1989). Orthographic depth and the interaction ofvisual and auditory processing in word recognition. Memory & Cog resented in memory through the first syllable only (Taft, nition, 17, 302-310. 1979, 1987; Taft & Forster, 1976). In English, too, mem FROST, R., KATZ, L., & BENTlN, S. (1987). Strategies for visual word ory representation oflonger words seems to incorporate recognition and orthographical depth: A multilingual comparison. Jour more than the first syllable, as is suggested by the lack nal ofExperimental Psychology: Human Perception & Performance, of differentiation between the first and the second sylla 13, 104-115. GARDNER, E. F., CALUS, J. C., MERWIN, J. c., & MADDEN, R. (1972). bles ofthree-syllable words under stressed and unstressed Stanford Test of Academic Skills. New York: Harcourt, Brace, conditions. Moreover, the results ofthe differential treat Jovanovich. ment of two-syllable nouns and verbs in English could GEVA, E., & RyAN, E. B. (1987, June). Linguistic and memory corre be an indication of the different memory representations lates ofacademic skills in first and second language. Paper presented of these two word types in memory. At this point, how at the meeting ofthe Canadian Psychological Association, Vancouver. GLUSHKO, R. J. (1979). The organization and activation oforthographic ever, these suggestions can only be speculative, and more knowledge in reading aloud. Journal ofExperimental Psychology: Hu research is needed to validate them. man Perception & Performance,S, 657-673. GOLDMAN, H. B., & HEALY, A. F. (1985). Detection errors in a task with articulatory suppression: Phonological recoding and reading. REFERENCES Memory & Cognition, 13,463-468. HATCH, E., POUN, P., & PART, S. (1974). Acoustic scanning and syn ADAMS, M. J. (1981). What good is orthographic redundancy? In O. J. tactic processing: Three reading experiments with first and second Tzeng & H. Singer (Eds.), Perception ofprint: Reading research in language learners. Journal of Reading Behavior, 6, 235-242. experimental psychology (pp. 197-221). Hillsdale, NJ: Erlbaum. HEALY, A. F. (1976). Detection errors on the word the: Evidence for ANDREWS, S. (1986). Morphological influences on lexical acceSS: Lexical reading units larger than letters. Journal of Experimental Psychol or nonlexical effects? Journal ofMemory & Language, 25, 174-188. ogy: Human Perception & Performance, 6, 403-409. BERNHARDT, E. B. (1986). Cognitive processes in L2: An examination HEALY, A. F. (1980). Proofreading errors on the word the: New evi ofreading behaviors. In J. Lantolf & A. Labarca (Eds.), Second lan dence on reading units. Journal ofExperimental Psychology: Human guage acquisition in the classroom setting (pp. 35-50). Norwood, NJ: Perception & Performance, 6, 403-409. Ablex. HEALY, A. F., CONBOY, G. L., & DREWNOWSKI, A. (1987). Charac BLACK, M., & BYNG, S. (1986). Prosodic constraints in lexical access terizing the processing units of reading. In B. K. Britton & S. M. in reading. Cognitive Neuropsychology, 3, 369-409. Glynn (Eds.), Executive control processes in reading (pp. 279-296). CARR, T. H., & POLLATSEK, A. (1985). Recognizing printed words: Hillsdale, NJ: Erlbaum. A look at current models. In D. Besner, T. G. Waller, & G. E. Mac HEALY, A. F., & DREWNOWSKI, A. (1983). Investigating the bound Kinnon (Eds.), Reading research: Advances in theory and practice aries of reading units: Letter detection in misspelled words. Journal (Vol. 5, pp. 1-82). Orlando, FL: Academic Press. ofExperimental Psychology: Human Perception & Performance, 9, CLARKE, M. A. (1979). Reading in Spanish and English: Evidence from 413-420. adult ESL students. Language Learning, 29, 121-150. JARED, D., & SEIDENBERG, M. (1990). Naming multisyllabic words. COLOMBO, L. (1992). Lexical stress effect and its interaction with fre Journal ofExperimental Psychology: Human Perception & Perfor quency in word pronunciation. Journal ofExperimental Psychology: mance, 16,92-105. Human Perception & Performance, 18,987-1003. JARVELLA, R. J., JOB, R., SANDSTROM, G., & SCHREUDER, R. (1987). 324 CHITIRI AND WILLOWS

Morphological constraints on word recognition. In A. Allport, D. G. port, D. MacKay, W. Prinz, & E. Scheerer (Eds.), Language per Mackay, W. Prinz, & G. Scheerer (Eds.), Language perception and ception and production (pp. 227-244). London: Academic Press. production (pp. 245-262). London: Academic Press. SCHINDLER, R. M. (1978). The effect ofprose context on visual search JASTAK, S., & WILKINSON, G. S. (1984). Wide Range Achievement for letters. Memory & Cognition, 6, 124-130. Test-Revised. Wilmington, DE: Jastak Associates. SEIDENBERG, M. S. (1985). The time course ofphonological code acti JOSEPH, B. D., & PHILIPPAKI-WARBURTON, I. (1987). Modem Greek. vation in two writing systems. Cognition, 19, 1-30. London: Croom Helm. SEIDENBERG, M. S. (1987). Sublexical structures in visual word recog JUST, M. A., & CARPENTER, P. A. (1980). A theory of reading: From nition: Access units of orthographic redundancy? In M. Coltheart eye fixations to comprehension. Psychological Review, 87, 329-354. (Ed.), Attention and performance XII: The psychology of reading JUST, M. A., & CARPENTER, P. A. (1987). The psychology ofreading (pp. 245-263). Hillsdale, NJ: Erlbaum. and language comprehension. Boston: Allyn & Bacon. SEIDENBERG, M. S. (1989). Reading complex words. In G. N. Carl KATSAITI, L. (1986). Remembering, forgetting, and cueing of text in son, & M. K. Tanenhaus (Eds.), linguistic structure in language pro monolinguals andbilinguals. Unpublished doctoral dissertation, Uni cessing (pp. 53-105). Dordrecht, The Netherlands: Kluwer Academic versity of Toronto. Publishers. KATZ, L., & FELDMAN, L. B. (1981). Linguistic coding in word recog SEIDENBERG, M. S., & MCCLELLAND, J. L. (1989). A distributed de nition: Comparison between a deep and a shallow orthography. In velopmental model of word recognition and naming. Psychological A. Lesgold & C. A. Perfetti (Eds.), Interactive processes in reading Review, 96, 523-568. (pp. 85-106). Hillsdale, NJ: Erlbaum. SELKIRK, L. (1980). On prosodic structure and its relation to syntactic KATZ, L., & FELDMAN, L. B. (1983). Relation between pronunciation structure. Bloomington: Indiana University Linguistics Club. and recognition ofprinted words in deep and shallow orthographies. SMITH, P. T., & GROAT, A. (1979). Spelling patterns, letter cancella Journal ofExperimental Psychology: Learning, Memory, & Cogni tion and the processing of text. In P. A. Kolers, M. E. Wrolstad, & tion, 9, 157-166. H. Bouma (Eds.), Processing ofvisible language (pp. 309-325). New KAY, J. (1987). Phonological codes in reading: Assignment of sub-word York: Plenum. phonology. In A. Allport, D. G. Mackay, W. Prinz, & E. Scheerer SPOEHR, K. T., & SMITH, E. E. (1973). The role of syllables in percep (Eds.), Language perception andproduction (pp. 181-196). London: tual processing. Cognitive Psychology,S, 71-89. Academic Press. SPOEHR, K. T., & SMITH, E. E. (1975). The role of orthographic and KU~ERA, H., & FRANCIS, W. N. (1967). Computational analysis of phonotactic rules in perceiving letter patterns. Journal ofExperimental present-day American English. Providence, RI: Brown University Psychology: Human Perception & Performance, 104, 21-34. Press. STANNERS, R. F., NEISER, W. P., HERNON, W. P., & HALL, R. (1979). LIMA, S. D., & POLLATSEK, A. (1983). Lexical access via an ortho Memory representation for morphologically related words. Journal graphic code? The basic orthographic syllabic structure (BOSS) recon of Verbal Learning & Verbal Behavior, 18, 399-412. sidered. Journal ofVerbal Learning & Verbal Behavior, 22, 310-332. TAFT, M. (1979). Lexical access via an orthographic code: The basic LUKATELA, G., SAVIC, M. D., OGNJENOVIC, P., & TURVEY, M. T. orthographic syllabic structure (BOSS). Journal of Verbal Learning (1978). On the relation between processing the Roman and the Cyril & Verbal Behavior, 18, 21-39. lic alphabets: A preliminary analysis with bi-a1phabetical readers. Lan TAFT, M. (1987). Morphographic processing: The BOSS re-emerges. guage & Speech, 21, 113-141. In M. Coltheart (Ed.), Attention andperformance XII: The psychol LUKATELA, G., & TURVEY, M. (1980). Some experiments on the Ro ogy of reading (pp. 265-279). Hillsdale, NJ: Erlbaum. man and Cyrillic alphabets. In J. F. Kavanagh & R. C. Venezky TAFT, M., & FORSTER, K. I. (1976). Lexical storage ofpolymorphemic (Eds.), Orthography, reading and dyslexia (pp. 227-247). Baltimore: and polysyllabic words. Journal of Verbal Learning & Verbal Be University Park Press. havior, 15, 607-620. LUKATELA, G., & TURVEY, M. T. (1990). Phonemic similarity effects TAYWR, I., & TAYWR, M. M. (1983). The psychology ofreading. New and prelexical phonology. Memory & Cognition, 18, 128-152. York: Academic Press. MACWHINNEY, B., & BATES, E. (Eds.) (1989). The cross linguistic study TOMBAIDES, D. E. (1986). Epitomi tis historias tis hellenikis glossas of sentence processing. Cambridge: Cambridge University Press. [Compendium of the history of the Greek language]. Athens: Or MACWHINNEY, B., BATES, E., & KLIEGL, R. (1984). Cue validity and ganismos Ekthoseos Didaktikon Vivlion. sentence interpretation in English, German, and Italian. Journal of TREIMAN, R., & CHAFETZ, J. (1987). Are there onset and rime-like units Verbal Learning & Verbal Behavior, 23, 127-150. in printed words? In M. Coltheart (Ed.), Attention and performance MAGOULAS, G. (1979). Introduction to Modem Greek phonology. XII: The psychology ofreaJiing (pp. 281-298). Hillsdale, NJ: Erlbaum. Athens: University of Athens Press. TREIMAN, R., GOSWAMI, U., & BRUCK, M. (1990). Not all nonwords MASSARO, D. W., TAYWR, G. A., VENEZKY, R. L., JASTRZEMBSKI, are alike: Implications for reading development and theory. Memory P., & LUCAS, P. A. (1980). Letter and wordperception. Amsterdam: & Cognition, 18, 559-567. Elsevier, North-Holland. TREIMAN, R., & ZUKOWSKI, A. (1988). Units in reading and spelling. MIRAMBEL, A. (1959). La langue grecque. Paris: Librairie Journal ofMemory & Language, 27, 466-477. C. Klincksieck. TURVEY, M. T., FELDMAN, L. B, & LUKATELA, G. (1984). The Serbo O'REGAN, K. (1979). Saccade size control in reading: Evidence for the Croatian orthography constrains the reader to a phonologically ana linguistic control hypothesis. Perception & Psychophysics, 25, lytic strategy. In L. Henderson (Ed.), Orthographies and reading 501-509. (pp. 81-90). London: Erlbaum. PATTERSON, K., & COLTHEART, V. (1987). Phonological processes in VAN ORDEN, G. C. (1987). A ROWS is a ROSE: Spelling, sound, and reading: A tutorial review. In M. Coltheart (Ed.), Attention andper reading. Memory & Cognition, 15, 181-198. formance XII: The psychology of reading (pp. 421-447). Hillsdale, VAN ORDEN, G. C., JOHNSTON, J. C., & HALE, B. L. (1988). Word NJ: Erlbaum. identification in reading proceeds from spelling to sound to meaning. PROCTOR, J. D., & HEALY, A. F. (1985). A secondary task analysis Journal ofExperimental Psychology: Learning, Memory, & Cogni ofa word familiarity effect. Journal ofExperimental Psychology: Hu tion, 14, 371-386. man Perception & Performance, 11, 286-303. RUBENSTEIN, H., LEWIS, S. S., & RUBENSTEIN, M. H. (1971). Evi NOTES dence for phonemic recoding in visual word recognition. Journal of Verbal Learning & Verbal Behavior, 7, 645-657. I. This older version was preferred because it was thought that the SCHEERER, E. (1987). Visual word recognition in German. In A. All- newer (1982) version contained information that is culture specific. WORD RECOGNITION IN ENGLISH AND GREEK 325

2. In an attempt to ensure comparability of test items we translated lent" in a technical sense. Their comparability lies in the fact that they these three tests back into English. There was a correspondence of70% covered the same concepts/meanings, using age appropriate vocabulary, between the original WRAT-R and PPVT-R English tests and the ver as judged by the researchers and teachers involved in the study. sion that was created after the Greek tests were retranslated into En 3. Out of the 270 content words (135 for each text) used in the ex glish. This less than perfect correspondence reflects a problem inher perimental passages, only 3 English words had more than one stress ent in translation-namely, that it is difficult to establish a one-to-one pattern (conduct, combat, and record). However, the correct pronun translation between languages because the same concept can be expressed ciation should be obvious from the sentence context. In Greek, there through several synonyms, especially in the absence of context. When were 6 words that could be pronounced in two ways, depending on the a context is present, as with the reading comprehension test ofthe Stan position ofthe stress. Given the fact that Greek stress is directly marked ford Achievement Test, equivalence of translation is often difficult to in the word, confusion over which pronunciation to choose should be establish because ofthe differing semantic and syntactic patterns across nonexistent. languages. In this case, while the meaning is the same, the form through which the meaning is conveyed may be different. Because of these limitations and because there are no Greek word (Manuscript received December 24, 1992; frequency lists, we cannot claim that the tests we used were "equiva- revision accepted for publication July 30, 1993.)